YOLO26s Finetuned on Brackish Underwater
Fine-tuned YOLO26s object detector on the Brackish Underwater benchmark dataset, trained and evaluated as part of DetectionBench -- a framework for reproducibly benchmarking modern object detectors with identical training recipes and evaluation metrics across multiple real-world datasets.
Usage
Install Dependencies
pip install ultralytics huggingface_hub
Load Model from Hugging Face
from huggingface_hub import hf_hub_download
from ultralytics import YOLO
weights = hf_hub_download(
repo_id="dronefreak/brackish-yolo26s",
filename="best.pt"
)
model = YOLO(weights)
Run Inference
results = model.predict(
source="image.jpg",
conf=0.25
)
results[0].show()
Performance
Evaluated on the Brackish Underwater test split, using DetectionBench's standard evaluation pipeline (detectionbench-evaluate).
| Metric | Score (%) |
|---|---|
| mAP@50 | 99.1 |
| mAP@50-95 | 85.77 |
| Precision | 99.45 |
| Recall | 97.83 |
| F1 Score | 98.63 |
| Parameters | 10.0M |
| FLOPs | 22.8B (at 640 px) |
Brackish Underwater Model Zoo
Every model DetectionBench has trained and evaluated on Brackish Underwater so far, for full transparency -- see DetectionBench for the smaller, curated comparison set used on the project README.
| Model | mAP@50 | mAP@50-95 | Precision | Recall |
|---|---|---|---|---|
| YOLOv8s | 99.3 | 85.65 | 99.36 | 98.53 |
| YOLOv8m | 99.19 | 86.04 | 98.69 | 98.71 |
| YOLO26s | 99.1 | 85.77 | 99.45 | 97.83 |
| YOLO26n | 98.95 | 83.66 | 98.17 | 97.23 |
| YOLO11x | 98.89 | 86.3 | 98.84 | 98.25 |
| YOLO11n | 98.87 | 83.27 | 98.76 | 96.69 |
| YOLO26m | 98.74 | 85.93 | 98.74 | 97.32 |
| YOLOv8n | 98.41 | 83.01 | 99.51 | 96.95 |
Per-Class Performance
| Class | mAP@50 | mAP@50-95 |
|---|---|---|
| crab | 99.5 | 94.53 |
| fish | 99.44 | 90.17 |
| jellyfish | 97.76 | 75.73 |
| shrimp | 99.5 | 78.22 |
| small_fish | 98.88 | 76.55 |
| starfish | 99.5 | 99.4 |
Dataset
This model was trained on Brackish Underwater. For the full dataset description, provenance, license, and citation, see the dataset card:
https://huggingface.co/datasets/dronefreak/Brackish
Classes
- crab
- fish
- jellyfish
- shrimp
- small_fish
- starfish
Training Configuration
| Setting | Value |
|---|---|
| Dataset | Brackish Underwater |
| Framework | Ultralytics YOLO |
| Training Toolkit | DetectionBench |
| Epochs (configured max) | 500 |
| Epochs (actually trained) | 500 |
| Early Stopping Patience | 100 |
| Batch Size | 32 |
| Image Size | 640 |
| Optimizer | auto |
| Initial Learning Rate | 0.001 |
| Seed | 0 |
Repository Contents
best.pt
results.csv
args.yaml
BoxPR_curve.png
BoxF1_curve.png
BoxP_curve.png
BoxR_curve.png
confusion_matrix.png
confusion_matrix_normalized.png
val_batch0_pred.jpg
brackish_yolo26s_showcase.jpg
README.md
Related Resources
- Brackish Underwater dataset card on Hugging Face
- DetectionBench -- reproducible benchmarks for modern object detectors on real-world datasets
Training Framework
This model was trained using DetectionBench, an open-source framework for benchmarking object detectors across multiple real-world datasets with a common pipeline.
Features include:
- A dataset-adapter registry for converting real-world datasets into a canonical format
- Identical training/evaluation recipes across model families (Ultralytics YOLO/RT-DETR, RF-DETR)
- Hardware profiling (latency, FPS, VRAM, parameters, FLOPs)
- One-command reproducibility via versioned Hydra configs
If you find this model useful, please consider starring the repository.
Known Limitations
- Severe class imbalance:
crab(34.6%) andsmall_fish(30.4%) account for roughly two-thirds of all annotated boxes in the training set, whileshrimp(1.46%) andjellyfish(1.82%) are rare -- per-class accuracy on the minority classes is measured on comparatively few examples. - Roughly 15% of images have no annotated objects at all (background-only frames, by design -- part of the dataset's varying-visibility setup, not a data-quality issue).
- Single-location, single-camera capture: all footage comes from one fixed camera 9 meters below the surface on the Limfjords bridge, Denmark -- generalization to other underwater cameras, locations, or water types (brackish vs. marine/freshwater) is untested.
- Two-hop provenance: this dataset was converted to YOLO format via a third-party Roboflow export, not sourced directly from the original annotated-video release; images are pre-resized to 1920x1080 by that export.
Citation
If you use this model in your research, please consider citing the dataset and the model architecture:
@InProceedings{pedersen2019brackish,
title = {Detection of Marine Animals in a New Underwater Dataset with Varying Visibility},
author = {Pedersen, Malte and Haurum, Joakim Bruslund and Gade, Rikke and Moeslund, Thomas B. and Madsen, Niels},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}
@article{jocher2026yolo26,
title={Ultralytics YOLO26: Unified Real-Time End-to-End Vision Models},
author={Jocher, Glenn and Qiu, Jing and Liu, Mengyu and Lyu, Shuai and Akyon, Fatih Cagatay and Kalfaoglu, Muhammet Esat},
journal={arXiv preprint arXiv:2606.03748},
year={2026}
}